Signalling by protein phosphatases and drug development: a systems-centred view

Authors


B. N. Kholodenko, Systems Biology Ireland, Conway Institute, University College Dublin, Belfield, Dublin 4, Ireland
Fax: +353-1-716-6856
Tel: +353-1-716-6919
E-mail: boris.kholodenko@ucd.ie

Abstract

Protein modification cycles catalysed by opposing enzymes, such as kinases and phosphatases, form the backbone of signalling networks. Although, historically, kinases have been at the research forefront, a systems-centred approach reveals predominant roles for phosphatases in controlling the network response times and spatio-temporal profiles of signalling activities. Emerging evidence suggests that phosphatase kinetics are critical for network function and cell-fate decisions. Protein phosphatases operate as both immediate and delayed regulators of signal transduction, capable of attenuating or amplifying signalling. This versatility of phosphatase action emphasizes the need for systems biology approaches to understand cellular signalling networks and predict the cellular outcomes of combinatorial drug interventions.

Abbreviations
Cdc14/25

cell division cycle 14/25 homolog

DUSP

dual-specificity phosphatase

EGF

epidermal growth factor

EGFR

epidermal growth factor receptor

ERK

extracellular signal-regulated kinase

Eya1/3

eyes absent homolog 1/3

FAK

focal adhesion kinase

HER2

human epidermal growth factor receptor 2

KSR1

kinase suppressor of Ras 1

MAPK

mitogen-activated protein kinase

MEK

mitogen-activated protein kinase

p120RasGRP

120 kD Ras-GTPase activating protein

PDGF

platelet-derived growth factor

PI3K

phosphoinositide 3-kinase

PRL-3

phosphatase of regenerating liver-3

PSP

Ser/Thr phosphatase

PTP

protein tyrosine phosphatase

PTP1B

protein-tyrosine phosphatase 1B

RTK

receptor tyrosine kinase

SHP1/2

SH2 domain-containing protein tyrosine phosphatase 1/2

SOS

son of sevenless

TK

tyrosine kinase

Introduction

Multiple external cues, including growth factors, cytokines and mechanical forces, activate plasma membrane receptors such as receptor tyrosine kinases (RTKs), G-protein-coupled receptors and other receptor families. This creates spatio-temporal phosphorylation patterns that are propagated through a interconnected network of signalling proteins and cascades. Frequently combined with intrinsic intracellular signalling, this complex network not only transmits but also processes and decodes the external information and gives rise to cellular responses. For instance, signals from different receptors are integrated through common targets and pathway cross-talk, such as in cell-cycle control [1,2] or mammalian target of rapamycin-mediated signalling [3]. Phosphatases play a vital role in cellular signalling by controlling both the network dynamics and spatial localization of phosphoproteins [4,5]. Although kinases have previously been the focus of scientific interest, protein tyrosine phosphatases (PTPs) and Ser/Thr phosphatases (PSPs) are becoming increasingly studied both as a research topic and as targets for drug development.

Phosphatases act as both immediate and delayed negative regulators of protein phosphorylation, and this often results in attenuation or termination of signal transfer. Constitutive phosphatase activities shape the initial phosphorylation profiles of receptors, phosphorylated adaptors, Ser/ Thr kinases and other signalling proteins after transient activation by growth factors or other signals. These immediate signalling responses develop very rapidly on a scale of second and minutes, and as discussed below, phosphatases can play a dominant role in determining the spatio-temporal behaviour of protein phosphorylation systems in the cell. Induced phosphatase activities often create negative feedback loops, which adapt cells to more permanent external stimulation, over a time scale of hours. Interestingly, inhibition of PTPs by reactive oxygen species whose production is induced by activated RTKs, such as epidermal growth factor receptor (EGFR), can create also positive feedback loops that facilitate lateral propagation of EGFR phosphorylation at the plasma membrane [6]. This multifaceted role of phosphatases in signalling control is illustrated by many examples, for instance the dual role of the SH2 domain-containing protein tyrosine phosphatase 2 (SHP2) in activation of extracellular signal-regulated kinase (ERK), as discussed in detail below [7–9].

Compared to kinases, progress towards understanding of the regulation of phosphatases was lagging behind due to technical challenges. For example, it is only possible to assay the activity of a given protein phosphatase in vitro if the relevant substrate has already been phosphorylated by a relevant protein kinase. In the human genome, the numbers of PTPs and RTKs are very similar, implying that the versatility and specificity of functions of these kinases and phosphatases may also be similar [10]. Although the catalytic subunits of PSPs have overlapping targets, the substrate specificity of PSPs is often achieved through their regulatory subunits [11,12]. Different regulatory and scaffolding subunits recruit a catalytic subunit to specific subcellular locations where different targets reside. Individual ternary PSP complexes assembled in these locations show differential catalytic activities and endow a particular PSP with its substrate specificity. In this review, we focus on how substrate specificity is controlled for phosphatases of the PSP family.

Historically, kinases have been major drug targets for treatment of cancer and other diseases. However, the versatility of phosphatase functions and their involvement in multiple feedback mechanism makes phosphatases attractive targets for future drug development. We discuss how PSPs are advancing to the forefront of drug development. To demonstrate the potential of systems biology approaches in facilitating selection of therapeutic targets, we develop a simplified mathematical model of the EGFR/SHP2 signalling pathway, and explore in silico phosphatase-based therapies versus receptor inhibition. Both theoretical and experimental studies focusing on understanding roles of phosphatases in controlling the spatiotemporal dynamics of signalling networks are discussed. We also show how phosphatase dynamics are regulated by the transcriptional machinery and how such transcriptional feedback loops control the entire signalling system in the context of mitogen-activated protein kinase cascades.

Phosphatases shape the temporal dynamics of signalling cascades

Signal transduction via cascades of phosphorylation/dephosphorylation cycles is a hallmark of cell signalling. The highly conserved mitogen-activated protein kinase (MAPK) cascades, which have been extensively studied, control a range of important physiological processes, including proliferation, differentiation and apoptosis [13,14]. MAPK cascades consist of three sequential levels, with phosphorylation and subsequent dephosphorylation, respectively catalysed by a kinase from the preceding level and a phosphatase at the given level.

The activity of signalling cascades such as the MAPK network may be characterized by a number of key features, notably the amplitude and duration of the signal output, both of which have a relevant physiological impact. Signal amplitude of MAPK activation exceeding a certain threshold was found to be a requirement for the proliferation of fibroblasts [15]. On the other hand, the duration of MAPK activity in PC12 cells dictates whether the cells proliferate or differentiate [16]. Moreover, rapid and transient MAPK activation in rat hepatocytes promoted G1/S cell-cycle progression, while prolonged MAPK activation inhibited this process [17]. By influencing different repertoires of target genes, the amplitude and duration of MAPK activation are critical in determining cell responses [16–19], and thus their quantitative description can be used to obtain insights into differential roles of the participating phosphatases and kinases in determining the cascade signalling outputs.

Theoretical analysis of signalling cascades without feedback loops has shown that the action of phosphatases outweighs that of kinases, exerting a dominant effect on the regulation of signal duration [5]. Kinases influence signal amplitude rather than duration, although phosphatases can also contribute to the regulation of signal amplitude. This is particularly apparent in weakly activated pathways where only a small proportion of the total kinase pool is phosphorylated. Under these conditions, signal duration is entirely determined by phosphatases, becoming prolonged at slow dephosphorylation rates. Interestingly, the position of a phosphatase within the cascade does not affect the extent to which it affects signal duration [5]. Mathematical studies on specific systems such as the ERK pathway have provided further support to these predictions [20,21]. In one such study utilizing normal rat kidney fibroblasts [20], the cells were arrested in G0 phase, and doubly phosphorylated ERK (ppERK) concentrations were measured following stimulation with epidermal growth factor (EGF) in the presence of increasing doses of an mitogen-activated protein kinase kinase (MEK) inhibitor [20]. Under these conditions, increasing MEK inhibition resulted in a decreased peak of a transient ERK activation, but had little effect on its duration. However, use of a PTP inhibitor led to a broader peak of ppERK, indicating a prolonged duration, consistent with model predictions [20]. These studies suggest that, in signalling pathways such as the MAPK cascade, where signal duration strongly determines cell fate, targeting phosphatases rather than kinases is a more viable strategy to control cell responses.

Dual-specificity phosphatases as rapid feedback inhibitors

As mentioned above, MAPK pathway signalling has been implicated in governing cell fate decisions. Diverse cellular events, such as proliferation, differentiation, migration and apoptosis, all require proper functioning of MAPK cascades. A puzzling aspect has been how one core module, such as the Ras/Raf/MEK/ERK pathway, can elicit cell responses as diametrically opposite as proliferation and differentiation. It has emerged over time that the answer to this question is that these pathways, which were traditionally thought to be linear cascades, are embedded in complex signalling networks of feedback interactions [14,22,23]. When a signal is relayed from the extracellular membrane via the MAPK pathway into the nucleus, a networked pathway allows for additional regulation either by integrating information from alternative co-activated and suppressed pathways or by facilitating self-regulation of the pathway by incorporating feedbacks. Although other classes of protein phosphatase, such as protein serine/threonine phosphatase 2A (PP2A), also have direct or indirect regulatory effects on the MAPK cascade, here we focus on how dual-specificity phosphatases (DUSPs) elicit feedback control in the context of the Ras/Raf/MEK/ERK pathways.

In response to extra- and intracellular signalling cues, cells induce regulatory feedbacks by two essentially different mechanisms. Either the activities of signal transducers are altered, or the protein concentrations of these transducers are changed. The activity changes are generally achieved post-translationally by altering modifications such as phosphorylations that occur rapidly and at multiple levels of the pathway. For instance, ERK alone can phosphorylate and inactivate several upstream signal transducers, including EGFR, son of sevenless (SOS), Raf-1 and MEK [14]. Such feedback controls mediated by post-translational modification occur almost immediately after the initial signal has been triggered. Furthermore, protein concentrations can be changed either by increasing protein degradation, such as depletion of EGFR [24], or by triggering a rapid transcriptional response, which can be induced on the time scale of minutes. Many genes that are strongly induced in this first transcriptional wave are direct regulators of upstream signalling, showing that biological systems exploit both post-translational and transcriptional feedbacks. Nevertheless, transcriptional feedbacks are inherently slower and more costly than post-translational ones, as it takes time and energy to induce substantial amounts of a nascent protein. Therefore, increasing the concentrations of feedback inhibitors has a delayed effect on the signalling cascade, and this inhibition is frequently sustained for longer periods. The increase in feedback inhibitor concentration allows the cellular system, which is initially very sensitive to extracellular cues, to adapt to the new environment. This is done by adjusting the threshold required for signalling, by reducing the signalling sensitivity, and by altering the dynamics of the response. The adaptation can involve reducing the receptor abundance [24], expressing specific antagonists, such as Sprouty [25] and MIG-6 [26], or inducing expression of phosphatases that dephosphorylate activating phosphorylation sites on signal transducers [27,28]. One class of phosphatases that is robustly induced upon activation of MAPK cascades are DUSPs [S. Keyes, unpublished results]. DUSPs are a sub-family of PTPs that bind MAPKs and dephosphorylate residues in their activation loop, leading to MAPK deactivation. Interestingly, DUSP activity is additionally regulated by the substrates, and binding to MAPK increases DUSP activity [30]. Additional regulation is achieved through post-translational modifications, as many DUSPs are themselves substrates of MAPK [31]. Therefore, inducible over-expression of DUSPs potently decreases MAPK activity and is considered to be part of the cellular feedback mechanism.

These feedback inhibitors are required to respond rapidly and with sufficiently high precision to changes in MAPK activity. Rapid turnover times are achieved through fast protein and mRNA degradation rates, which are hallmarks of these feedback regulators [32]. This allows rapid transcriptional regulation of the protein, which in turn permits accurate and reliable tuning of the signalling response.

Induced expression of phosphatases reduces the dose–response sensitivity and the signalling output, but can also fundamentally change the dynamics of the response. For instance, NIH-3T3 cells show rapid and sustained phosphorylation of the downstream ERK1/2 kinases following stimulation with platelet-derived growth factor (PDGF) [33]. Interestingly, the pathway activation appears to be self-sustaining, as MAPK activity persists even when PDGF is washed out after the initial stimulus. Under these conditions, ERK1/2 activity is not linearly related to the PDGF input concentration. Incremental increasing PDGF concentrations do not lead to incremental increases in MAPK signalling, but instead result in a switch-like, ‘all-or-nothing’ activation above a certain threshold, similar to an ultrasensitive activation. Although DUSPs are induced by PDGF, their expression does not dramatically affect ERK1/2 activation dynamics [33]. Although post-peak MAPK activity is reduced, the sustained MAPK activation dynamics persist. Intriguingly, the behaviour of the system can change dramatically if cellular DUSP expression is substantially increased by pre-exposing NIH-3T3 cells to low PDGF concentrations. Re-stimulating these pre-conditioned NIH-3T3 with increasing PDGF concentrations dramatically changes the dose–response sensitivity of the MAPK activity. The previously ultrasensitive system now displays a linear relationship between PDGF input and MAPK phosphorylation output. The transformation of a switch-like ultrasensitive response to a graded response illustrates the flexibility and adaptability of the cellular signalling network, in this case mediated by an inducible phosphatase acting as a feedback regulator [33]. Interestingly, DUSPs also shape the dynamics of mitogenic responses. This was elegantly demonstrated by utilizing a cell line that expresses a rapidly inducible Ras isoform that harbours an oncogenic activating mutation and constitutively stimulates the downstream Raf/MEK/ERK pathway [34]. Following expression of mutated Ras, ERK activity initially overshoots the input but is rapidly reduced after 30 min, resulting in a sharp activity peak. Importantly, after 30 min, the Ras input and ERK phosphorylation output show a linear dose–response relationship. A mathematical model of this system showed that, in order to mimic this behaviour, ERK activity has to react initially in an ultrasensitive manner, but this input–output relationship subsequently changes with DUSP expression. Thus, these findings confirm the results obtained in NIH-3T3 cells, further demonstrating that expression of DUSPs affects the amplitude, dose–response relationships and temporal dynamics of MAPK activation. Acting in this manner as rapid feedback regulator, DUSPs tightly control MAPK activity. However, computational studies suggested that if DUSP-mediated feedback is too strong, it can also result in oscillations [22,35]. Such oscillations have been experimentally observed in the Fus3 MAPK pathway that is responsible for regulating the mating pheromone response in Saccharomyces cerevisiae [36]. Strong correlation between the oscillatory Fus3 activation peaks and periodic formation of additional mating projections suggests important physiological role for these oscillations. Experiments and mathematical modelling found that transcriptional induction of the MAPK phosphatase Msg5 and the negative regulator of G-protein signalling Sst2 are required for maintenance of these oscillations [36].

Recent evidence further indicates that feedback control by DUSPs can shape the dynamics of the MAPK response differentially, depending on the cellular compartment [18]. In MCF7 cells, the MAPK pathway responds to heregulin treatment with a robust and sustained activation of ERK in the cytoplasm. Surprisingly, when ERK phosphorylation is monitored in the nuclear fraction, the sustained cytoplasmic ERK signal is translated into a transient response. This is true even if phosphorylated ERK is normalized by the amount of total nuclear ERK, taking nuclear/cytoplasmic shuttling into account. Intriguingly, knockdown of nuclear DUSPs by siRNAs is sufficient to transform heregulin-induced ERK phosphorylation within the nucleus from transient to sustained. Therefore, it appears that the difference between nuclear and cytoplasmic ERK dynamics may be due to the presence or higher expression and induction of specifically nuclear-localized DUSPs.

Overall, it is becoming clear that phosphatase-mediated deactivation of MAPK pathways is used by the cell to control and regulate all aspects of signalling, be it the duration, amplitude or localization of the signal. As a result of this additional control, the system can react with high adaptability and flexibility to changing and diverse environmental stimuli.

Spatial separation of phosphatases and kinases can give rise to phosphoprotein gradients

Cells are three-dimensional structures, and the spatial regulation of protein activities is important for many physiological processes, including cell division, motility and migration. In addition to their roles in temporal dynamics, phosphatases control the spatial behaviour of protein phosphorylation systems within the cell. When a protein that is phosphorylated at the plasma membrane spreads solely by diffusion, dephosphorylation mediated by a cytosolic phosphatase results in a steep gradient of phosphorylation signal [4,37]. This gradient is characterized by high concentrations of phosphorylated protein close to the membrane and low concentrations in the cell interior, with the decay profile being almost exponential if the phosphatase is far from saturating condition [38]. Interestingly, how fast the gradient is terminated depends only on the diffusion coefficient and the apparent first-order rate constant of the phosphatase but not the kinase. This result suggests that the localization and catalytic activity of phosphatases may play important roles in determining spatial signalling gradients in cells. Experimental evidence for such activity gradients is accumulating, including the small GTPase Ran [39], the yeast MAPK Fus3 [40], the phosphatase protein-tyrosine phosphatase 1B (PTP1B) [41], aurora B kinase [42] and the yeast protein kinase Pom1 [43]. Further work investigating spatial signal propagation in simplified signalling cascade models revealed similar constraints to those found for the temporal responses. For activation signals to readily spread from the cell membrane into the cell interior, the Vmax/Km ratios for the phosphatases must be much lower than those for the kinases [44,45].

Drug development targeting protein phosphatases

Kinases have been the major targets of drug discovery efforts [46]. This is partly because kinases were thought to have dominant control over signalling systems, while phosphatases were considered less important counterparts of the kinases with unclear involvement in cell-fate decisions, mainly as consequence of specific technical challenges in the study of phosphatases. As discussed above, this view is becoming obsolete, as phosphatases may have a significant influence in shaping the spatio-temporal dynamics of signalling pathways, thereby affecting cell-fate decisions. Emerging systems biology approaches that combine mathematical modelling with quantitative experimentation can facilitate understanding of the network complexity and selection of therapeutic targets. Although development of drugs targeting protein phosphatases is progressing, only a few such drugs have progressed into clinical trials, and the degree of success of therapies targeting phosphatases is yet to be determined. The main efforts focus on the treatment of diabetes, Alzheimer’s disease and cancer. Here we summarize some of the strategies that have been used to target various classes of protein phosphatases and how systems biology can be used to develop better treatments that target phosphatases (see [47–50] for more detailed reviews).

Targeting protein tyrosine phosphatases

The protein tyrosine phosphatases are characterized by the presence of the conserved sequence (H/V)C(X)5R(s/T) in the active site [51]. Of more than 100 PTPs, only a few are considered to be possible therapeutic targets [10]. For instance, although some PTPs behave as oncogenes, an RNAi screen against 107 PTPs has shown that, in HeLa cells, knockdown of only four PTPs had a negative effect on cell survival, while knockout of 28 PTPs increased cell survival [52]. This screen shows that activation of some of these PTPs can potentially be used as an anti-tumour treatment. However, the development of drugs that specifically target a particular PTP is complicated by two factors: (a) the high level of homology of the phosphatase domains of various PTPs, and (b) the fact that the targeted sequences are highly charged, and many of the compounds developed cannot cross the membrane [48]. To increase the specificity, non-homologous neighbouring domains of the active site may also be targeted. In addition, the cell permeability of drugs can be increased by chemical manipulation [48].

Drugs targeting PTPs displaying oncogenic behaviour are in various phases of development. The proteins being targeted are PTP1B, SH2 domain-containing protein tyrosine phosphatase 2 (SHP2), cell division cycle 25 homolog (Cdc25), cell division cycle 14 homolog (Cdc14), phosphatase of regenerating liver-3 (PRL-3) and eyes absent homolog 1/3 (Eya1/3) [49]. Mutations of these proteins or changes in the level of expression appear to play a role in cancer and autoimmune diseases. For example, PTP1B is a negative regulator of the insulin receptor [53], and there is evidence that inhibition of this phosphatase increases sensitivity to insulin, making PTP1B a very attractive target for the treatment of obesity and diabetes [54]. Interestingly, orthovanadate was originally developed as a drug to treat diabetes, long before it was known that it inhibits PTPs. PTP1B may also positively regulate human epidermal growth factor receptor 2 (HER2) [55], subsequently activating several proteins in the downstream EGF signalling network, such as Src [55,56] and 120 kD Ras-GTPase activating protein (p120RasGRP) [47]. Therefore, it may also be a potential therapeutic target for the treatment of breast cancer. PTP1B inhibitors have been developed using various approaches: the most recent generation are bidentate difluoromethylphosphonates that are designed to target the active site and a secondary substrate binding region close to the catalytic pocket [57]. These inhibitors bind PTP1B with higher affinity than other related PTPs, and are being modified to increase their cell permeability [58] but have not gone into clinical trials yet. Two PTP1B inhibitors, ertiprotafib and trodusquemine, have advanced into clinical trials for the treatment of obesity and diabetes, although the second-phase clinical trial for ertiprotafib was discontinued due to lack of efficacy [49,59]. A phase I clinical study of trodusquemine is currently being performed [60]. Another PTP that has been targeted is SHP2, a phosphatase that contains two SH2 domains [61] and is considered a bona fide oncogene that regulates cell progression and migration by modulating ERK1/2 and focal adhesion kinase (FAK) signaling [7]. SHP2 is required for full activation of ERK, and impaired SHP2 activity was responsible for the surprising finding that activating mutations in EGFR failed to fully induce ERK activation [62]. Activating mutations of SHP2 have been identified in patients with various leukaemias, solid tumours and in several germline mutation syndromes, such as Noonan and Leopard syndromes [63]. Gain-of-function mutations in the N-SH2 domain impair auto-inhibition of the PTP domain, and usually result in increased signalling from Ras and other oncogenes such as Src, and a general increase in the downstream signal from various growth factor receptors [48]. Several SHP2 small molecule inhibitors have been produced and are in various phases of development (for a detailed review, see [49]). One of the bigger problems in development of these inhibitors is that SHP2 shows high homology with SHP1, another PTP that acts as tumour suppressor. Thus, the SHP2 inhibitors must not inhibit SHP1, or must have a higher affinity for SHP2 at the administration dose (Fig. 1). To date, no SHP2-specific inhibitors have advanced to clinical trials; however, a dual SHP1/2 and PTP1B inhibitor is currently in clinical trials in combination with interferon α treatment. This inhibitor appears to be well tolerated and augments immunological responses [64].

Figure 1.

 SHP1 and SHP2 are phosphatases that play opposite roles in the regulation of signalling pathways. (A) In cancer, a SHP2-activating mutation or SHP2 stimulation by oncogenic signals results in activation of oncogenic pathways such as the Ras/ERK and Src pathway. (B) Inhibitors that specifically target SHP2 or have higher affinity for SHP2 than for SHP1 inhibit of these pathways, shifting the balance towards tumour suppression. FAK, focal adhesion kinase; PI3K, phosphoinositide 3-kinase.

Targeting protein serine/threonine phosphatases

As mentioned above, protein Ser/Thr phosphatases (PSPs) include a variety of proteins with more than 30 catalytic subunits that interact with various regulatory and structural subunits. The PSPs actually comprise several families: the phospho-protein phosphatases, metallo-dependent protein phosphatases and Asp-based enzymes. However, the phospho-protein phosphatases are responsible for the majority of serine and threonine dephosphorylation [65]. These proteins have been shown to play an important role in the regulation of various biological functions, in close relationship with tyrosine kinases [66]. The PSPs are key regulators of kinase activity, and their functional de-regulation has been observed in various pathologies such as cancer and Alzheimer’s disease. Of the many members of the PSP family, PP2A has recently become a target for drug development, specifically in the context of cancer therapy. Several isoforms of PP2A act as bona fide tumour suppressors that negatively regulate mitogenic signals [67]. This phosphatase is also de-regulated in various types of cancer such as breast, lung and melanoma [68,69]. Inhibition of PP2A is necessary for the transformation and tumour progression of various cancers. Studies involving both mutation and loss of expression of all PP2A subunits have been performed [50]. In addition to de-regulation of PP2A subunits, the PP2A inhibitory proteins SET and Phasin 1 (PHAP-1) have also been linked to various malignancies. For instance, SET is over-expressed in BCR/ABL-driven leukaemias [70], and PHAP-1 has been shown to be related to aberrant phosphorylation of tau protein in Alzheimer’s disease [71]. In light of these observations, various drugs that restore normal PP2A activity are being studied. For example, sodium selenate decreases the tau protein phosphorylation levels in mice and is currently is under intense study [72]. In the context of cancer, the rationale for developing PP2A-targeting drugs is that restoration of the phosphatase enzymatic activity would result in inhibition of the transforming signal caused by oncogene expression. So far the best known activator of PP2A is FTY720, a structural analogue of sphingosine that has been approved for the treatment of multiple sclerosis [73]. In animal models of leukaemia, FTY720 has been shown to increase the rate of survival with few toxic side-effects [50,74], indicating that PP2A activation may be a safe strategy in cancer treatment. Although still in the early stages, these studies demonstrate that targeting PSPs such as PP2A is a potentially useful therapeutic strategy; however, the complex spatio-temporal regulation evident within these phosphatase networks suggests that further understanding is required to generate the sensitivity and specificity essential for therapeutic applications.

Regulatory subunits in the spatio-temporal control of phosphatases

Although phosphatases such as DUSPs ensure spatio-temporal regulation of pathway activity through tight transcriptional control and internal localization sequences, regulation of other phosphatases, such as the diverse family of PSPs, represents an entirely different type of control. In contrast to monomeric phosphatases, the specificity and control of PSP activity (with the notable exception of the monomeric PP2Cs of the metallo-dependent protein phosphatase family) are mediated by formation of a multi-component complex containing a catalytic subunit and a regulatory subunit. In some cases, the assembly is facilitated by a scaffolding subunit, resulting in a trimeric complex [75].

Although the substrate specificity of kinases has been established on the basis of a linear motif recognition surrounding the phosphorylated amino acid, sites of PSP-directed dephosphorylation do not display significant sequence similarity [76]. Instead, substrate specificity is achieved through docking of the phosphatase complex at a site distant to the dephosphorylated amino acid [12,76]. Consensus motifs for regulatory subunit docking sites have been established for some prominent members of the PSP family, including PP1 and PP2B (calcineurin), but not for PP2A, for which multiple interactions and post-translational modifications play a role in directing catalytic activity [77].

Numerous studies have demonstrated that specificity within the human PP2A network is achieved through differential assembly of heterotrimeric complexes from the genomic repertoire of two catalytic subunits (PP2aCα/β), two scaffolding subunits (PR65α/β) and at least 15 known regulatory subunits encoded by four separate gene families (termed B, B′, B′′ and B′′′) [78]. By exploiting this differential assembly mechanism, a multitude of individual heterotrimeric complexes can be produced, exerting control over a wide array of cellular processes [79,80]. Furthermore, post-translational modifications also play a significant role in temporal regulation of the PP2A assembly at both the catalytic and regulatory subunits [77]. Although incorporation of B and some B′ regulatory subunits is inhibited upon Src-mediated phosphorylation of PP2aC, methylation of PP2aC may be required for incorporation of B and possibly B′ subunits [77]. Phosphorylation of regulatory subunits also contributes to this temporal regulation in a kinase- and subunit-specific manner [77,81]. A further layer of spatial regulation is added to these heterotrimeric complexes through the variety of localization sequences within the regulatory subunits, limiting PP2A activity to specific subcellular locations [77,79].

A prime example of a network regulated by PP2A in this complex manner is that governing activation of ERK following growth factor stimulation. Within this network, PP2A can act at multiple levels to promote either activation or inhibition of ERK, depending upon the site of PP2aC recruitment, a process that is controlled by various regulatory subunits (Fig. 2). Upon growth factor stimulation, PP2aC is recruited to the kinase suppressor of Ras 1 (KSR1)/Raf1/MEK complex through the B family member PR55α/δ, where it is required for Raf1 activation via dephosphorylation of the inhibitory S259 site [82] and also of 14-3-3 binding sites within Raf1 and KSR1 [83]. However, PP2aC acts via B’ family member PR61β/δ to directly inhibit ERK [84] and also indirectly promotes tyrosine dephosphorylation of Shc through an unidentified regulatory subunit [85]. Additionally, PP2aC also inhibits Ras-independent ERK activation by dephosphorylating c-Src upon interaction with an alternative B family member PR55γ [86].

Figure 2.

 Schematic representation of sites of PP2A activity within the network of ERK activation. Ligand-mediated activation of receptor tyrosine kinases (RTK) at the plasma membrane leads to activation of the classical Ras/Raf/MEK kinase pathway, leading to phosphorylation of ERK (pathway components shown in white, activating phosphorylations indicated by black arrows). Individual heterotrimeric PP2A complexes (shown in grey) containing a catalytic subunit (PP2aC), scaffolding subunit (PR65) and various regulatory subunits are spatially separated within the network based upon specific interaction between network components and each regulatory subunit. Activating dephosphorylation of network components by PP2aC is indicated by a grey arrow, and inhibitory dephosphorylation is indicated by a round-ended grey arrow.

Multifaceted regulatory and combinatorial assembly mechanisms such as these present a significant challenge for experimental characterization of the global PP2A network, a vital step when considering PP2A as a therapeutic target. Many studies have focused on individual complexes and their specific dephosphorylation targets, providing extensive data on how these complexes act in isolation [77–81]; however, little is known about regulation of PP2A at the network level.

Recent studies utilizing systems-level approaches have begun to yield significant advances in this field. At one level, mathematical modelling has allowed characterization of specific PP2A heterotrimers, the abundance of which was too low to measure experimentally [87,88], whilst interactomics-based studies are beginning to piece together the PP2A network as a whole [89]. A recent study utilized mass spectrometry-based interactomics to investigate the whole network of interactions occurring across PP2A catalytic, scaffolding and regulatory subunits [89]. This study confirmed the simultaneous existence of a large pool of heterogeneous heterotrimeric PP2A complexes, and placed these into distinct modules characterized by the presence of regulatory subunits linked to specific cellular processes. Intriguingly, this study highlighted the underlying complexity of the PP2A network by hinting at the existence of higher-order complexes containing proteins not previously associated with this network. Additionally, it also suggested the utilization of PP2A regulatory subunits by other PSP families, demonstrating evolutionary divergence of the human PP2A network from that of lower eukaryotes.

Although systems biology approaches are just starting to unravel the complex interactions and modifications involved in regulation of the PP2A network, building of systems-level protein–protein interaction networks such as this will lay the foundation for further studies examining the dynamic behaviour of these systems.

How can systems biology accelerate drug development targeting protein phosphatases?

Regulation of protein phosphorylation is crucial for many biological processes, and de-regulation of kinases that catalyse phosphorylation leads to the onset of various diseases. This prompted the development of agents that target tyrosine kinases (TKs) years before phosphatases were considered as therapeutic targets. Most TK-targeting drugs were developed to inhibit specific kinases and were initially used as single agents, based on the concept of oncogene addiction. However, apart from imatinib, most of the inhibitors that are used in the clinic have failed as single agents and are given in combination with other treatments. This is due to an inherent biological redundancy whereby various TKs have overlapping targets, and inhibition of a single kinase is not sufficient to restore the normal intracellular phosphoprotein levels. Another problem is non-specific inhibition of other kinases, reducing the benefits of inhibiting a particular kinase. Hopefully, the lessons learned from the development of TK inhibitors can speed up the development of drugs targeting phosphatases.

The use of mathematical models can help in the identification of appropriate targets, and predict the efficiency, course of treatment and drug combinations that may have a therapeutic effect [46]. Although use of mathematical models is still very limited in drug development, there are already examples that show how systems biology can be useful for drug development and clinical application [90–92]. For example, a mathematical model of the PI3K/AKT pathway was used in combination with clinical data to identify new biomarkers that can help to decide which patients would benefit from treatment with PI3K inhibitors and RTK inhibitors [93]. Similar approaches have been used to predict the drug inhibition profile for the NF-κB pathway [94], or to identify optimal therapeutic targets for activation of p53 [95].

However, the complex nature of phosphatase biology makes it extremely difficult to predict off-target effects and determine which patients will benefit from treatment with particular pharmacological agents. Furthermore, as mentioned, one of the challenges for the development of agents that target PTPs is their selectivity, which is hindered by the high homology among members of the PTP family. The use of mathematical models can help to identify ‘therapeutic windows’ that will inhibit a given PTP without resulting in deleterious inhibition of other PTPs. For instance, inhibitors with a higher affinity for SHP2 than for SHP1 could be used at doses that only affect SHP2. Furthermore, it is also likely that there will be synergistic effects of combinations of phosphatase-targeting drugs and TK inhibitors. These synergistic drug combinations may also be predicted using mathematical models.

Phosphatase-targeted therapies suggested by a computational model of the EGFR pathway

Although efforts in targeting protein phosphatases for therapeutic purposes are already underway, this endeavour may be significantly accelerated if guided by mathematical modelling and systems approaches to cell signalling. These methods can help in identification of suitable targets and prediction of potential drug treatments and the efficiency of combined therapies. As an illustrative example, we used a simplified mathematical model that incorporates the EGFR/ERK pathway and the tyrosine phosphatase SHP2 (the EGFR/SHP2 system) to explore alternative therapeutic strategies to inhibit ERK activation mediated by over-expressed EGFR. Our model predicts that, within certain cancer cell contexts, suppression of ERK activation by targeting the phosphatase SHP2 may be more effective than targeting the receptor.

SHP2 has been reported to a have dual regulatory role [62]. It negatively regulates phosphorylation of RTKs (e.g. EGFR and insulin receptor) and adaptor proteins (e.g. insulin receptor substrate and Grb2-associated binder 1 [9]). However, SHP2 has strong positive effect on Ras activation, facilitating full activation of the ERK. This positive effect is related to plasma membrane recruitment of SHP2 through binding to phosphorylated tyrosine residues on the insulin receptor substrate and Grb2-associated binder 1 scaffolds. SHP2 then subsequently dephosphorylates multiple docking sites involved in binding and membrane recruitment of the GTPase-activating protein for Ras (RasGAP), enhancing Ras activity [7,8]. To account for the action of SHP2 on the downstream Raf-1/MEK/ERK cascade, we constructed an ordinary differential equations (ODE)-based model that extends our previously established EGRF network model [19] to describe the SHP2/Ras/ERK pathway. The reactions involved in the model are illustrated in the scheme presented in Fig. 3. Rate equations and parameter values are given in Tables S1–S3. Briefly, in the EGFR/SHP2/Ras/ERK model, signal transduction is initiated by EGF binding to the extracellular domain of monomeric EGFR (designated R in the kinetic scheme, reaction 1, Fig. 3). This causes dimerization and autophosphorylation of EGFR (reactions 2, 3 and 25), which is subsequently dephosphorylated by several phosphatases (reactions 4 and 26). To account for the combinatorial complexity of phosphorylation of various sites on EGFR and the fact that SHP2 specifically dephosphorylates the sites involved in RasGAP binding, we use an approximate ODE description by considering two separate forms of phosphorylated EGFR, designated RP1 and RP2 (Fig. 3) (see [96–98] for more rigorous approaches to reduce combinatorial complexity of signal transduction networks). We assume that binding of proteins to these two tyrosine residues is statistically independent. The RP2 form mediates RasGAP binding and is dephosphorylated by active SHP2 that has bound to RP1 (reactions 5-8 and 27). The adaptor proteins Shc and Grb2 bind competitively to the RP1 form, and Grb2 also binds to phosphorylated Shc (reactions 13-18). The Grb2–SOS complexes bound to either EGFR or the EGFR–Shc complex catalyse conversion of RasGDP to RasGTP (reaction 28), and the reverse transition is catalysed by RasGAP bound to the RP2 form of EGFR (reaction 29). Activated Ras subsequently activates the Raf/MEK/ERK cascade (reactions 30–39).

Figure 3.

 Kinetic scheme of EGFR/SHP2 signalling mediated by adapter and target proteins. Numbering of individual steps is arbitrary.

Although this dynamic model is not comprehensive, it can be exploited to compare alternative therapies that perturb distinct classes of targets. Many cancer cell types have elevated expression of EGFR [62]. Inhibiting EGFR using an EGFR inhibitor such as gefitinib appears to be a preferred treatment to suppress ERK pathway activity. However, many patients show reduced gefitinib sensitivity, and new treatments that can overcome gefitinib resistance are required. Using a mathematical model, we compared two therapies that target either SHP2 (a phosphatase-based therapy) or EGFR (a kinase-based therapy).

This model shows that, in normal cells, characterized by low physiological EGFR levels, EGF stimulation induces transient responses of active Ras (RasGTP) and ppERK, consistent with the experimental observations (Fig. 4A,B). In cancer cells, EGFR over-expression often leads to sustained RasGTP and ppERK responses, which is also reflected by the model predictions (Fig. 4C,D) [99]. Importantly, predictive simulations performed using the model suggest that, when the SHP2 level is also high in cancer cells, inhibition of SHP2 better suppresses active ERK and RasGTP levels compared to EGFR inhibitors (Fig. 4C,D). Additionally, the model predicts that combining the two inhibitors in a dually targeted therapy further decreases RasGTP and ppERK, thereby enhancing the efficiency of the treatment (Fig. 4C,D). For comparison purposes, the model assumed that both EGFR and SHP2 inhibitors reduce the concentrations of SHP2 and EGFR by 40% of their pre-treatment levels. Interestingly, increasing the dosage of both inhibitors not only further suppresses the Ras/EGFR pathway, but also increases the efficacy of SHP2-based therapy over the EGFR-based therapy (data not shown).

Figure 4.

 Comparison of two molecularly targeted therapies for the EGFR/SHP2 signalling system as illustrated in Fig. 3. (A,B) Time-course concentrations of active RasGTP (A) and doubly phosphorylated ERK (B) in response to EGF stimulation for a normal cell with low EGFR level (100 nm). (C,D) Time-course concentrations of RasGTP (C) and doubly phosphorylated ERK (D) in response to EGF stimulation in cancer cells, characterized by up-regulated levels of EGFR (800 nm). Time-course data are also included for the presence of an EGFR inhibitor (red line), an SHP2 inhibitor (blue line) and combined treatment with both EGFR and SHP2 inhibitors (dashed purple line). Details of model equations and parameter values are given in Tables S1–S3.

These simulations demonstrate that, under certain conditions, targeting SHP2 can be a more viable strategy in suppressing ERK activation than targeting a RTK receptor. They also highlight the emerging concept that the design of signal transduction therapies requires understanding of the underlying mechanisms that control aberrant signalling patterns and pathological cell traits. As we have demonstrated, systems biology approaches can reveal these hidden regulatory patterns and open fresh avenues for drug discovery.

Concluding remarks

The perception of phosphatases as enzymes whose role is solely to counteract kinases in linear signalling pipelines from receptors to target genes has been replaced by an emerging concept of a complex kinase/phosphatase network that is tightly regulated through a multitude of negative and positive controls by feed-forward and feedback loops. Phosphorylation and dephosphorylation of multiple tyrosine, serine and threonine residues on signal transducers results in dramatic changes to their activities, leading to specific alterations in cellular phenotypes. This complex combinatorial nature of cellular signalling highlights the requirement for systems biology methods to understand the roles of phosphatases in determining signalling dynamics and their targeting in drug development.

We have begun to rationalize how the intricate network circuitries can determine the spatio-temporal signalling kinetics to precisely translate them into specific biological responses. We show that phosphatases can act as both immediate and delayed controllers of signal processing, and how the effects of this regulation can be negative or, surprisingly, positive in amplifying cellular responses. In addition, mathematical models have shown that phosphatase rather than kinase activities predominantly control the response time of distinct signalling processes and phosphorylation/dephosphorylation cascades. The prominent role of phosphatases in shaping the spatial profiles of signalling activities within a cell has recently been revealed by both computational and experimental studies. Systems biology models are emerging as a novel tool to accelerate drug development. As phosphatase targeting comes to the forefront of drug development, there is the need to assess the systems consequences of drug-induced changes in phosphatase activities. Here we illustrate how computational modelling can help predict the outcomes of drug therapies targeting various cellular processes. Further development of systems-level approaches will facilitate the selection of proper treatments for specific pathological conditions.

Acknowledgements

We thank Walter Kolch (Director of Systems Biology Ireland, Conway Institute. University College Dublin, Belfield, Dublin 4) for stimulating discussions and reading the manuscript. This work was supported by Science Foundation Ireland (grant number 06/CE/B1129) and National Institutes of Health grant GM059570.

Ancillary